CMU-Perceptual-Computing-Lab / openpose

OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation
https://cmu-perceptual-computing-lab.github.io/openpose
Other
30.41k stars 7.8k forks source link

Question | OpenPose and tf-pose estimation #847

Closed Rutulpatel7077 closed 5 years ago

Rutulpatel7077 commented 5 years ago

I love OpenPose. Thanks for the great library.

I am absolute beginner.

My question is why generated keypoints (JSON file) are different in both OpenPose and tensorflow-pose-estimation ?

What is the main difference?

is there any way to compare both keypoints results ?.

Tensorflow JS Pose Estimation example

{
  "score": 0.32371445304906,
  "keypoints": [
    { // nose
      "position": {
        "x": 301.42237830162,
        "y": 177.69162777066
      },
      "score": 0.99799561500549
    },
    { // left eye
      "position": {
        "x": 326.05302262306,
        "y": 122.9596464932
      },
      "score": 0.99766051769257
    },
    { // right eye
      "position": {
        "x": 258.72196650505,
        "y": 127.51624706388
      },
      "score": 0.99926537275314
    },
    ...
  ]
}

OpenPose example

{
    "version":1.1,
    "people":[
        {
            "pose_keypoints":[
                376.034,327.151,0.948575,443.19,430.934,0.785305,372.936,433.979,0.736717,363.816,620.328,0.793385,253.757,571.352,0.766581,516.383,427.911,0.766405,482.803,608.042,0.749979,376.084,501.212,0.844809,443.176,736.191,0.510191,376.046,919.351,0.851897,400.538,1075.03,0.653059,553.102,739.226,0.522832,498.057,986.485,0.842563,0,0,0,357.694,305.745,0.968881,400.494,305.648,0.976137,0,0,0,449.234,311.902,0.751298
            ],
            "face_keypoints":[

            ],
            "hand_left_keypoints":[

            ],
            "hand_right_keypoints":[

            ]
        }
    ]
}
lvcoc commented 5 years ago

please,can you tell me how to get JSON with tf-pose estimation

stale[bot] commented 5 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

aligajani commented 5 years ago

Just a different output because posenet uses TF.js and OpenPose Caffe.

2429581027 commented 4 years ago

Hello, your question has been solved, how do you calculate the score?

Rutulpatel7077 commented 4 years ago

@2429581027 Score is just an accuracy of the prediction

andreibirladeanu commented 2 years ago

please,can you tell me how to get JSON with tf-pose estimation

I wrote this function which parses the Human tf-pose object as there wasn't any way to get that data in the tensorflow 2.0 implementation https://github.com/andreibirladeanu/phd_code/blob/master/openpose_tf/extract_skeleton.py